Multiprocessor networks with small buffers: theory and simulation

There is a rise of interest in the performance of the networks with small buffers in recent years. Analysis showed that increase in buffer size does not improved the performance in wormhole routing significantly [1--4]. Moreover, latest research proved that use of smaller buffers does not decrease the link utilization [5] for Internet routers. In interconnection networks, buffer depth of 5 packets had been shown to offer optimal performance for optical packet-switched clockwork routing [6] The use of smaller buffer sizes give some advantage in speed and possibilities to use SRAM or OPS (Optical Packet Switching). We present theoretical models and numerical results for performance of a multiprocessor network modeled as a ring of routers with small finite buffers (size 1 and 2), and with local processors that generate messages with rate λ per time slot and two output ports/buffers. Simulation results are also obtained for networks with torus topology (four output ports, total buffer capacity of 8 and 20 messages). The average queue lengths and average latency are obtained. The results show that the model of independent queues, which is valid for networks with infinite buffers, is still applicable for small λ, but breaks down, which violates the Jackson theorem.

[1]  Leonard Kleinrock,et al.  Theory, Volume 1, Queueing Systems , 1975 .

[2]  Lev B. Levitin,et al.  Performance model of a multiprocessor interconnection network: theory and simulation , 2007, SpringSim '07.

[3]  Lev B. Levitin,et al.  Modeling of Latency and Saturation Phenomena in Interconnection Networks , 2007, 2007 12th IEEE Symposium on Computers and Communications.

[4]  Guido Appenzeller,et al.  Sizing router buffers , 2004, SIGCOMM '04.

[5]  Donald F. Towsley,et al.  Congestion Control for Small Buffer High Speed Networks , 2007, IEEE INFOCOM 2007 - 26th IEEE International Conference on Computer Communications.

[6]  Leonard Kleinrock,et al.  Queueing Systems: Volume I-Theory , 1975 .

[7]  Seok Ho Chang,et al.  Performance analysis of a finite-buffer discrete-time queue with bulk arrival, bulk service and vacations , 2005, Comput. Oper. Res..

[8]  Lev B. Levitin,et al.  Analysis and simulation of a model of multiprocessor networks , 2007 .

[9]  Ahmed Louri,et al.  Adaptive Channel Buffers in On-Chip Interconnection Networks— A Power and Performance Analysis , 2008, IEEE Transactions on Computers.

[10]  Lev B. Levitin,et al.  Interconnection networks with heterogeneous activity or finite buffers: beyond Jackson's theorem , 2008, CNS '08.

[11]  Baltasar Beferull-Lozano,et al.  Efficient routing with small buffers in dense networks , 2005, IPSN 2005. Fourth International Symposium on Information Processing in Sensor Networks, 2005..

[12]  Tim Roughgarden,et al.  Routers with Very Small Buffers , 2006, Proceedings IEEE INFOCOM 2006. 25TH IEEE International Conference on Computer Communications.

[13]  Nutthapol Asadathorn,et al.  A decomposition approximation for assembly‐disassembly queueing networkswith finite buffer and blocking , 1999, Ann. Oper. Res..

[14]  E. Bravi,et al.  Optical packet-switched interconnect based on wavelength-division- multiplexed clockwork routing , 2007 .

[15]  Michel Mandjes,et al.  Large Deviations for Small Buffers: An Insensitivity Result , 2001, Queueing Syst. Theory Appl..

[16]  Sudhakar Yalamanchili,et al.  Interconnection Networks: An Engineering Approach , 2002 .